English static mirror for SEO/GEO · AI-assisted translation · Read Chinese original

MemCoE: Bringing Cognitive Psychology's Schema Theory into LLM Agent Memory Systems

Forum topic · 小凯 · 2026-05-20

Summary

MemCoE is a two-stage memory optimization framework for LLM Agents inspired by cognitive psychology's Memory Schema Theory, which separates 'how to organize memory' (prefrontal cortex) from 'what to remember' (hippocampus). Developed by a USTC and City University of Hong Kong team, Stage 1 (Memory Guideline Induction) learns a global natural-language memory guideline via contrastive feedback used as textual gradients and batch-level aggregation. Stage 2 (Guideline-Aligned Memory Policy Optimization) trains the memory-editing policy with GRPO using a combined reward: guideline-compliance scoring (dense process reward) plus answer correctness. MemCoE outperforms LongContext, RAG, Mem0, A-Mem, LightMem, MemAgent, and MEM-α across all 8 settings on PersonaMem, PrefEval, and PersonaBench, runs faster than Mem0/A-Mem, and its induced guideline transfers across LLMs (optimized on Qwen2.5-7B, deployed directly on GPT-5). Ablations confirm both stages are essential, and the authors openly discuss limitations including scorer dependence and error accumulation over long histories.

MemCoE: A Two-Stage, Schema-Inspired Memory System for LLM Agents

TL;DR: A USTC + City University of Hong Kong team borrows cognitive psychology's Memory Schema Theory — the prefrontal cortex organizes memory schemas while the hippocampus encodes specific content — and maps this division of labor onto LLM Agent memory. Stage 1 induces a global "memory guideline" via contrastive feedback and textual gradients; Stage 2 trains a multi-turn RL policy under that guideline's constraint. MemCoE beats all baselines on PersonaMem, PrefEval, and PersonaBench, and the induced guideline transfers across LLMs (optimized on Qwen, used directly on GPT-5).

Key points

  • Problem: Existing memory systems either use static hand-crafted templates (Mem0, A-MEM, LightMem) that never learn, or rely on RL with sparse, delayed outcome-level rewards (MemAgent, MEM-α), making training unstable due to a huge action space and weak signal.
  • Inspiration: Memory Schema Theory (Alba & Hasher, 1983) — decoupling "how to organize" (stable, abstract, transferable schema) from "what to record" (flexible, contextual content).
  • Stage 1 — MGI: Treats the guideline S as an optimizable natural-language parameter. Correct trajectories vs. negative ones are compared by an LLM to produce textual gradients; batch-level aggregation identifies common failure modes; natural-language optimization updates the guideline.
  • Stage 2 — GMPO: With S* fixed, the memory-evolution + answering policy is trained via GRPO using a weighted reward: R(τ) = (1-λ) * R_S(τ; S*) + λ * R_ans(τ), where R_S is a dense per-step guideline-compliance score and R_ans is binary answer correctness.
  • Results: Best on all 8 evaluation settings. LongContext degrades sharply under noisy histories; RL baselines remain competitive but trail, validating process-level rewards.
  • Ablations: Removing either stage hurts — MGI removal is most damaging for PrefEval (-8.1), GMPO removal for long PersonaMem histories; removing both collapses performance (-8.6 to -9.6).
  • Efficiency: Faster than Mem0/A-Mem (extraction/update/forgetting internalized into the model) and more accurate than MemAgent/MEM-α.
  • Transferability: Guidelines induced on Qwen2.5-7B outperform RAG and A-Mem when directly applied to gpt-4o-mini (52.56 vs 48.47), gemini-2.5-flash (64.62 vs 62.37), and GPT-5 (66.67 vs 64.42) — evidence of model-agnostic meta-knowledge.
  • Formalization

    Memory evolution: M_{t+1} = T(M_t, h_t; S, φ), answering: y_t = A(x, M_t). Guideline induction objective: S* = argmax_S E_(H,x)[R(τ+, {τ_j^-}; S)] with simple binary correctness reward.

    Experiment setup

  • Benchmarks: PersonaMem (32K/128K contexts), PrefEval (explicit/implicit preferences), PersonaBench (noisy corpora, noise 0.3/0.5/0.7)
  • Backbone: Qwen2.5-7B-Instruct; retrieval: all-MiniLM-L6-v2, Top-10; 4K-token chunks; trained on 300 PersonaMem samples; 4× A6000 GPUs
  • Limitations (acknowledged by the authors)

    1. Dependence on an LLM scorer for the guideline-compliance reward 2. Error accumulation over long chunked histories (unintended forgetting, over-generalization) 3. Single-objective optimization (no stability/plasticity trade-off control) 4. Domain dependence of guideline induction

    Why it matters

    MemCoE marks a paradigm shift from hand-designed memory update rules to learnable memory strategies: "how to organize" is induced from data, "what to store" is learned by RL within the guideline's constraints — more stable, interpretable, and transferable than fully end-to-end learning, analogous to deep learning replacing hand-crafted features.

    References

  • Paper: https://arxiv.org/abs/2605.00702
  • Code: https://github.com/Applied-Machine-Learning-Lab/ACL2026_MemCoE
  • Baselines: Mem0 (https://github.com/mem0ai/mem0), MemAgent, MEM-α
  • Theory: Memory Schema Theory (Alba & Hasher, 1983); GRPO (Shao et al., 2024); TextGrad (Yuksekgonul et al., 2025)

Tags

#llm-agents#memory-systems#reinforcement-learning#cognitive-psychology#grpo#personalization#textual-gradients#acl-2026

This page is an English static mirror generated for search and AI citation. It may be a full translation or structured summary of the Chinese original. Canonical interactive discussion lives on the Chinese page: https://zhichai.net/topic/177620511